Class-Specific, Top-Down Segmentation

Size: px
Start display at page:

Download "Class-Specific, Top-Down Segmentation"

Transcription

1 Class-Specific, Top-Down Segmentation Eran Borenstein and Shimon Ullman Dept. of Computer Science and Applied Math The Weizmann Institute of Science Rehovot 76100, Israel {boren, Abstract. In this paper we present a novel class-based segmentation method, which is guided by a stored representation of the shape of objects within a general class (such as horse images). The approach is different from bottom-up segmentation methods that primarily use the continuity of grey-level, texture, and bounding contours. We show that the method leads to markedly improved segmentation results and can deal with significant variation in shape and varying backgrounds. We discuss the relative merits of class-specific and general image-based segmentation methods and suggest how they can be usefully combined. Keywords: Grouping and segmentation; Figure-ground; Top-down processing; Object classification 1 Introduction A major goal of image segmentation is to identify structures in the image that are likely to correspond to scene objects. Current approaches to segmentation mainly rely on image-based criteria, such as the grey level or texture uniformity of image regions, as well as the smoothness and continuity of bounding contours. In this work we describe a segmentation method that is guided primarily by high-level information and the use of class-specific criteria. The motivation for using such class-based criteria to supplement the traditional use of image-based criteria in segmentation has two parts. First, it stems from the fact that although recent image-based segmentation algorithms provide impressive results, they still often fail to capture meaningful and at times crucial parts. Second, evidence from human vision indicates that high-level, class-based criteria play a crucial role in the ability to segment images in a meaningful manner (e.g. [11],[10],[9],[8]), suggesting that the incorporation of such methods will help improve the results of computer vision segmentation algorithms. Figure 1 demonstrates some of the major difficulties encountered by imagebased segmentation algorithms. An appropriate segmentation algorithm should group together the dark and light regions of the horse (left), and separate the man from the horse, despite the grey level similarity (right). Figure 2 shows This research was supported by the Israel Ministry of Science under the Scene Teleportation Research Project and by the Moross Laboratory at the Weizmann Institute of Science. A. Heyden et al. (Eds.): ECCV 2002, LNCS 2351, pp , c Springer-Verlag Berlin Heidelberg 2002

2 110E. Borenstein and S. Ullman the results obtained when applying a state of the art image-based segmentation algorithm [15] to these and other images. The segmentations demonstrate some of the inherent difficulties of an image-based approach, including the splitting of object regions and the merging of object parts with background regions. These shortcomings are due to unavoidable ambiguities that cannot be solved without prior knowledge about the object class at hand. This paper presents an approach that addresses these difficulties, in which a simple representation of object classes in memory is used to guide the segmentation process, leading to markedly improved segmentation of images containing familiar objects (Fig. 3). The overall structure of the paper is as follows. Section 2 briefly reviews past approaches. Section 3 provides an overview of our approach, and how information about object shapes is represented in memory and used for segmentation. Section 4 describes the approach in detail: describe the segmentation criteria used by our method. 4.4 describes the algorithm that segments class-images according to these criteria. Section 5 shows results, Sect.6 contains a final discussion and conclusions. Fig. 1. Segmentation difficulties: the same object can contain markedly different regions (left), while neighboring objects may contain regions that are similar in color and texture (right). 2 Brief Review of Segmentation Approaches Most of the current approaches to segmentation rely primarily on image-based criteria, such as color, grey level, or texture uniformity of image regions (e.g. [16],[4],[2]); the smoothness and continuity of their bounding contours (e.g. [7]); or a combination of these (e.g. [6]). The region-based approaches merge and split image regions according to specific criteria. Merging approaches recursively merge similar regions (e.g. [14],[1]). Divide & Conquer approaches recursively split regions into distinct sub-regions (e.g. [12],[15]). Contour-based approaches emphasize the properties of region boundaries, such as continuity, smoothness, length, curvature, and shape. Somewhat closer to our goal is the work on deformable templates (e.g. [18]), where the template parameters are used to match a model with an object image. This approach usually assumes approximate initial correspondence between

3 Class-Specific, Top-Down Segmentation 111 Fig. 2. Typical results of low-level segmentation. Objects are broken into sub-regions, and some of these sub-regions are merged with the background. (More results can be seen at doron/software/ncuts/results/) Fig. 3. Low-level vs. class-specific segmentation of horse images. Top: input images (170x120pixels). Middle: low-level segmentation into sub-regions, as given by the normalized cuts algorithm [15]. Bottom: figure-ground segmentation map given by the algorithm described in this paper (input resolution reduced to 40x30).

4 112 E. Borenstein and S. Ullman the model and its image and therefore it can be used as a final stage in our segmentation but not as the main segmentation process. Recent psychological studies of image segmentation by humans indicate strong effects of prior familiarity with specific shapes on image segmentation. For example, a number of behavioral studies have shown that subjects are more likely to regard a familiar region as figure than a less familiar region, indicating that object recognition facilitates segmentation [11],[10]. Developmental studies [9],[8] show that figure-ground segregation in four month-old infants is also affected by the presence of already familiar shapes. 3 Overview of the Approach The general thrust behind our class-based segmentation approach is to use known shape characteristics of objects within a given class to guide the segmentation process. The main difficulty in this approach stems from the large variability of shapes within a given class of objects. We would like to somehow capture the common characteristics of a set of shapes within a class, for example horse images, and then use this information to segment novel images. To address this problem our approach uses a fragment-based representation of object classes. Similar fragments have been used in the past for object classification [13][17] but not for segmentation. Given an image containing a certain object, we use fragments previously extracted from images of the same object class to produce a consistent cover of the novel object. This cover defines a figure-ground map that associates each pixel in the input image with the likelihood of belonging to an object or background. (Fig. 4). The construction of an object by fragments is somewhat similar to the assembly of a jigsaw puzzle, where we try to put together a set of pieces such that their templates form an image similar to a given example. A common strategy is to start with the easiest pieces (e.g. corners) and proceed by connecting additional pieces that match in shape, color, edges, texture, etc. In some cases, as information accumulates along this process, pieces must be replaced: locally these pieces provide good matches, but the global structure adds constraints that reject the local matches. The next several sections describe our segmentation algorithm in detail. We first describe the fragment representation how the class fragments are represented in memory, and how they are extracted from sample images. 3.1 Fragment Representation in Memory In this section we describe the fragment-based representation used for segmentation. The goal of this representation is to cover as closely as possible the images of different objects from a given class, using a set of more primitive shapes. We therefore need to identify useful building blocks, a collection of components that can be used to identify and delineate the boundaries of objects in the class. To find such common components we look for image fragments that are strongly correlated with images containing the desired object class they show a high

5 Class-Specific, Top-Down Segmentation 113 Fig. 4. Overview of the approach. Given an input image containing an object from a given class (in this case, a horse) we use class fragments stored in memory to produce a cover of the target image. Each fragment consists of a template and a figure-ground label. The cover defines the figure-ground segmentation. similarity measure with regions from images containing this desired object class but not with others. Based on the Neyman-Pearson decision theory, optimal fragments can be defined as fragments with maximal frequency (hit rate) within the class, subject to the constraint that the frequency of false detection in non-class images (false alarms) does not exceed a fixed limit (e.g. [3]). Our search for optimal fragments therefore proceeds in three stages. Stage 1 starts from a set of training images divided into class images (C) and non-class images (NC) and then generates a large number of candidate fragments. We simply extract from the images in C a large number of rectangular sub-images, these sub-images can vary in size and range from 1 50 to 1 7 of the object size. In stage 2, which is the crucial step, we compare the distribution of each fragment in the class and non-class training images. For a given fragment F i, we measure the strength of the response S i in C and NC. S i is defined in a standard way: we correlate F i with each image I in C and NC (normalized correlation) and take the maximum value over I. To reach a fixed level of false alarms α in non-class images we determine a threshold θ i for F i by the criterion: p(s i >θ i NC) α (1) This has the advantage of automatically fixing an optimal detection threshold for each fragment. In stage 3 we order the fragments by their hit rate p(s i >θ i C) and select the K best ones where K determines the size of the fragment set. To be used for segmentation, we add two factors to each fragment: a figureground label and a reliability value. The figure-ground label marks each pixel

6 114 E. Borenstein and S. Ullman in the fragment as figure or ground. The figure label is learned at present by comparing the fragment to the source image in the data base from which it was derived. We assume that in this limited set the figure-ground information has been estimated, for example by relative motion of the figure with respect to background. An alternative is to label pixels in the fragment according to their grey level variability in the class database. Figure pixels are similar across images and have low-variability, while background pixels show high variability. The reliability of a fragment measures the extent to which the fragment is class-specific, measured by its hit rate p(s i >θ i C). These two factors are essential to our segmentation process, as will be later demonstrated. Fig. 5. Fragment representation in memory: Each fragment is represented by its grey level template (left) and figure-ground label (right). 4 Segmentation by Optimal Cover The main stage of the class-based segmentation algorithm consists of covering an image with class-based fragments and using the cover to delineate the figure boundaries. To accomplish this we seek an optimal cover of the image in terms of the fragments. A cover is an assignment of fragments to positions in the image I, with each fragment being defined as either present in the image along with its designated position p i in the cover or absent from the image: I { cover fi =1,p i =(x i,y i ) i=1...k f i =0,p i = φ (2) Given a cover, we can compute the quality of the cover, which is a function of the individual match of fragments with the image, the consistency of the cover, and the reliability of the participating fragments. The following sections describe each of these factors in turn and an algorithm to find a locally optimal cover according to these criteria. 4.1 Individual Match The individual match measures the similarity between fragments and the image regions that they cover. We use a similarity measure that combines region correlation with edge detection. This combination which is not commonly used,

7 Class-Specific, Top-Down Segmentation 115 is designed to meet the special requirements of segmentation as illustrated in Fig. 7. A key feature of our fragment representation is the elimination of background noise. Using the figure-ground label it is possible to exclude background pixels from the similarity measure, thereby reducing background noise. Since this exclusion results in the loss of the contour features, an edge detector is added to capture the transition from figure to ground (Fig. 7). The edge presence is computed by using a normalized correlation between the figure-ground boundary and edges in the image (Fig. 6). The similarity measure s i (p, I) between fragment F i at image position p =(x, y) and an image I is defined in (3) and consists of two factors. The first factor N cor is the standard normalized correlation between the fragment and the image region, restricted to template pixels labeled as figure. The second term E dge is the edge detector response. s i (p, I) =w N cor (p, I) Object Pixels +(1 w) E dge (p, I) (3) Fig. 6. The edge template (right) is derived from the boundary of the figure-ground label. The response to this template is the edge part E dge in (3). 4.2 Consistency In covering the image by shape fragments, the fragments should not only provide good local matches, but should also provide a consistent global cover of the shape. We therefore use a consistency criterion for the cover in the segmentation process. Since the fragments are highly overlapping, we define a consistency measure c ij between a pair of overlapping fragments F i and F j that is proportional to the fraction of pixels labeled consistently by the two fragments (4). The maximum term in the denominator prevents overlaps smaller than a fixed value µ ij from contributing a high consistency term. This value is set to 1 10 of the maximum possible overlap size between the two fragments. Fig. 8 demonstrates two cases of overlapping horse fragments : one in which all the pixels are labeled consistently and one in which some of the pixels are labeled inconsistently. c ij = # Consistent Overlapping Pixels max(total Overlap,µ ij ) (4)

8 116 E. Borenstein and S. Ullman Fig. 7. The examples show why a combination of templates and boundary similarity are useful for segmentation. A1 An image of a horse and a template of its back. A2 - Changes in the background reduce the similarity measure (measured using both the figure and ground parts of the template), causing the template to be more similar to another image region. B1 In this example the template fit was measured using the figure part of the template only. This reduces background effects but also results in the loss of boundary information leading to inaccurate matches. C1,C2 Adding an edge detector to the similarity measure yields a more stable similarity measure resulting in an accurate placement of the fragment regardless of background noise. Fig. 8. Consistent (left) and inconsistent (right) cover by overlapping fragments. 1,4: the fragments, placed over the images, 2,3: figure-ground assignment of the fragments. Figure pixels are marked white, background pixels are grey. The inconsistent region is marked in black.

9 Class-Specific, Top-Down Segmentation Fragment Reliability The final criterion in determining an optimal cover is that of fragment reliability. Similar to a jigsaw puzzle, the task of piecing together the correct cover can be simplified by starting with some more reliable fragments, or anchor fragments, and then proceeding with less reliable fragments that are consistent with the initial cover (Fig. 9). Reliable fragments typically capture some distinguishing features of the shapes in the class and are unlikely to be found anywhere else. A fragment s reliability is therefore evaluated by the likelihood ratio between the detection rate and the false alarm rate. As explained, we set the minimal threshold such that the false alarm rate does not exceed α. We can therefore express this ratio using the detection rate and α: r i = p(s i >θ i C) detection rate = p(s i >θ i NC) α (5) Fig. 9. Reliable fragments guide the covering. Reliable fragments are used first (middle), and subsequently completed by less reliable ones (right). 4.4 The Cover Algorithm A cover on an image by shape fragments determines a segmented figure. Among all possible covers we seek a cover (2) that maximizes the three criteria above, namely, individual match quality (3), consistency (4) and reliability (5). These three factors are therefore combined in the cover score: cs = r i s i f i + 1 β ij f i f j (6) λ i i,j } {{ } } {{ } Individual match Consistency and reliability The first term combines the match quality and reliability of the fragments, and the second penalizes inconsistent overlapping pairs. We define the interaction term β ij between overlapping fragments as (c ij β) (r i s i +r j s j ) where c ij is the pairwise consistency defined above (4) and β a global constant that determines the magnitude of the penalty for insufficient consistency. For non-overlapping

10 118 E. Borenstein and S. Ullman pairs β ij is defined as 0. The contribution of a single fragment F k in this expression is obtained by summing up all the terms in (6) for which i = k. When a fragment contribution is negative, the score is improved by removing the fragment from the cover. Negative contribution indicates poor consistency of the fragment with other fragments and can happen only when c kj <β λ for at least one j. In our implementation β = 0.65,λ = 0.1. The algorithm is iterative, but a small number of iterations (typically 2-3) are used. It is described in the Appendix, but the main stages are summarized next. At each stage, a small number M of good candidate fragments are identified. A subset of these M fragments, that maximally improve the current score, are selected and added to the cover. In addition, existing fragments that are inconsistent with the new match are removed. We use a small number of candidates (M = 15) that allows us to check all 2 M subsets and select the one with the highest score. The algorithm is guaranteed to converge to a local maximum since the score is bounded and increases at each iteration. To initialize the process, we select a sub-window within the image with the maximal concentration of reliable fragments. The similarity of all the reliable fragments is examined at 5 scales at all possible locations giving a complexity which is linear in the number of reliable fragments, the number pixels of each image scale, and the number of scales. Given this information it is possible to pick the most reliable window and use the matched fragments inside as the initial M candidates for the cover. If the combined evidence from the reliable fragments falls below a classification threshold, the process terminates without producing a cover. In a system containing multiple classes that compete for segmentation (rather than just horse images), the class with the highest evidence will initiate the cover. 5 Experiments We tested the algorithm on a database containing horse images. A bank of 485 fragments was constructed from a sample library of 41 horse containing images of size (40x30) for which the figure-ground information was manually segmented. For each fragment we estimated p(s i C) and p(s i NC) by measuring the distribution of the fragments similarity measure with 193 low-resolution images of horses and 253 low-resolution images of non-horses. Using these estimated distributions, the fragments were assigned their appropriate threshold and classified to 146 reliable and 339 non-reliable fragments. The algorithm was then tested on 176 novel horse images (40x30 pixels). Examples are shown in Fig. 10 and compared to the results of a normalized-cuts segmentation algorithm [15] tested on the same images but with higher resolution (170x120). The algorithm obtains high-quality segmentations of figure from background for a variety of images. The algorithm can deal successfully with shape variations using a fixed repertoire of fragments extracted from the training set. The generalization to novel shapes is based in part on the use of multiple alternative fragments for the same object region, and in part in the flexibility in the fragments arrangement. We also compared qualitatively the agreement between the figure regions produced by the algorithm and the figure region judged by humans. This can be expressed

11 Class-Specific, Top-Down Segmentation 119 by evaluating the ratio r = S F S F where F is the human-segmented figure and S is the algorithm s segmented figure. F, S are the sizes of F, S and the size of the entire image I = 1. The maximal value of r is r = 1 obtained only for perfect segmentation. The average score for the current algorithm was r = The normalized-cuts algorithm, segmenting the images into two segments figure and ground, gives much lower average score (r =0.31). The last value can also be compared with random segmentation (where S pixels are chosen randomly to be figure), which gives an average of r =0.23. The most problematic figure regions were the horse legs, where variability is high. The initialization step was the most time consuming (about 30 seconds), where the similarity of every reliable fragment was examined on 5 different scales of the target image. Usually the algorithm converged after 2-3 iterations, with each iteration taking a few seconds, giving a total time of about 40 seconds per target image. We used Matlab 6.0 program on a Windows-NT, Pentium-600Mhz platform. The complexity of the algorithm is linear in the number of scales, number of fragments, and size of the fragment bank. 6 Discussion The approach proposed in this paper emphasizes the role of high-level information or class-specific criteria in image segmentation. We present a class-specific segmentation method that successfully addresses ambiguities inherent to segmentation schemes based exclusively on image-based criteria. Segmentation is obtained by covering the image with a subset of class-specific fragments and using this cover to delineate the figure boundaries. These fragments serve as class-specific shape primitives, or building blocks, and are used to handle a large variety of novel shapes within the class. The fragments are represented by a template together with a figure-ground label, and a reliability score. The last two factors in this fragment representation are novel characteristics of our approach and are essential for segmentation. The figure-ground label is essential for: (a) constructing the figure-ground segmentation map; (b) defining a robust similarity measure that reduces background noise; and (c) defining the consistency between fragments. The fragment reliability enables us to detect key fragments that serve as initial anchors in the cover construction. In order to construct a complete cover we use both reliable and less reliable fragments. The less reliable fragments are constrained by the consistency relations of the cover and can therefore be used to complete the segmentation of difficult regions. Compared with other schemes that apply segmentation and classification in sequence, in our scheme the two processes are intimately linked. The initial evidence from reliable class fragments is used to select the most likely class that serves to initiate the cover, and the final classification decision may depend on the segmentation result. The algorithm results in the segmentation of images into two regions, figure and ground, in contrast with image-based segmentation algorithms that usually segment the image into multiple regions. The extraction of objects from these

12 120E. Borenstein and S. Ullman Fig. 10. Segmentation results arranged in 3 groups of 4 rows. First row in each group: input images. Second row: results obtained from low-level segmentation. Third row: class-based segmentation to figure and ground. Fourth row: segmentation superimposed on the input images (boundary in red).

13 Class-Specific, Top-Down Segmentation 121 images is challenging because objects are often broken into sub-regions, and some of these sub-regions merge with the background. The main difficulty faced by the current algorithm lies in covering highly variable parts, such as the horse s legs and tail. This difficulty is due in part to image processing limitations and the use of relatively low-resolution images. In the future we intend to use higher resolution images or a pyramid of image segments at different scales. Compared with the class-based segmentation, traditional image-based segmentation methods have two advantages. First, when they detect the correct figure boundaries, they can determine these boundaries with higher accuracy since they are guided directly by image discontinuities. Second, image-based algorithms are general and do not require class-specific information. The relative merits of class-specific and image-based segmentation methods suggest that they can be usefully combined into an integrated scheme. For example, image-based segmentation can be used to identify salient regions and direct class-based segmentation to these regions. At the final segmentation stage, figure boundaries produced by class-based segmentation could be refined by image-based methods (e.g. [5]) resulting in a robust and accurate delineation of object boundaries that cannot be achieved by either method alone. 7 Appendix: The Class-Based Segmentation Algorithm Pre-processing: For all reliable fragments F i, compute s i (p j,i) for all image positions p j and all scales (see(3)). Set s(p j ) = max i [r i s i (p j )] (best fragment at p j ). For each image window W : Pick in W at most M positions p j with maximal values of s(p j ). Define the score of the window W score = p j s(p j ). Initialization: Choose window W max to be the window with maximal W score together with its fragments. These fragments compose the initial covering candidates B cand. Set the current cover B cover to be empty. Choosing the new covering fragments: With all fragments (F j B cover ) fixed to f j = 1, assign the candidate fragments F i B cand to f i = {0, 1} such that (6) is maximized. Add to B cover all candidate fragments with f i = 1 and remove all the fragments that reduce the score of (6). Use B cover to construct the figure ground segmentation map. If B cover did not change, then stop.

14 122 E. Borenstein and S. Ullman Updating the candidate fragments sets: From all fragments in W max select the M with the highest score (individual match and consistency with B cover ). Go to choosing the covering fragments. References 1. K. Cho and P. Meer. Image segmentation from consensus information. Computer Vision and Image Understanding: CVIU, 68(1):72 89, J.M.H. du Buf, M. Kardan, and M. Spann. Texture feature performance for image segmentation. Pattern Recognition, 23: , R. Duda, P. Hart, and D. Stork. Pattern classification, A.K. Jain and F. Farrokhnia. Unsupervised texture segmentation using gabor filters. Pattern Recognition, 24: , M. Kass, A. Witkin, and D. Terzopoulos. Snakes: Active contour models. International Journal of Computer Vision, 1: , T. Leung and J. Malik. Contour continuity in region based image segmentation. In Fifth Euro. Conf. Computer Vision, Freiburg, Germany, D. Mumford and J. Shah. Boundary detection by minimizing functionals. In IEEE Conf. on Computer Vision and Pattern Recognition, San Francisco, A. Needham. Object recognition and object segregation in 4.5-month-old infants. Journal of Experimental Child Psychology, 78:3 24, A. Needham and R. Baillargeon. Effects of prior experience in 4.5-month-old infants object segregation. Infant Behaviour and Development, 21:1 24, M.A. Peterson. Object recognition processes can and do operate before figureground organization. Current Directions in Psychological Science, 3: , M.A. Peterson and B.S. Gibson. Shape recognition contributions to figure-ground organization in three-dimensional displays. Cognitive Psychology, 25: , M. Pietikainen, A. Rosenfeld, and I. Walter. Split and link algorithms for image segmentation. Pattern Recognition, 15(4): , E. Sali and S. Ullman. Combining class-specific fragments for object classification. In Proc. 10th British Machine Vision Conference, volume 1, pages , E. Sharon, A. Brandt, and R. Basri. Fast multiscale image segmentation. In Proc. IEEE Conference on Computer Vision and Pattern Recognition, pages 70 77, South Carolina, J. Shi and J. Malik. Normalized cuts and image segmentation. In Proc. IEEE Conf. Computer Vision and Pattern Recognition, pages , A. Tremeau and N. Borel. A region growing and merging algorithm to color segmentation. Pattern Recognition, 30, No. 7: , S. Ullman, E. Sali, and M. Vidal-Naquet. A fragment based approach to object representation and classification. In Proc. of 4th international workshop on visual form, Capri, Italy, A. Yuille and P. Hallinan. Deformable templates. In A. Blake and A. Yuille, editors, Active Vision, pages 21 38, MIT press, 1992.

Learning to Segment. Eran Borenstein and Shimon Ullman

Learning to Segment. Eran Borenstein and Shimon Ullman Learning to Segment Eran Borenstein and Shimon Ullman Faculty of Mathematics and Computer Science Weizmann Institute of Science Rehovot, Israel 76100 {eran.borenstein,shimon.ullman}@weizmann.ac.il Abstract.

More information

Combining Top-down and Bottom-up Segmentation

Combining Top-down and Bottom-up Segmentation Combining Top-down and Bottom-up Segmentation Eran Borenstein Faculty of Mathematics and Computer Science Weizmann Institute of Science Rehovot, Israel 76100 eran.borenstein@weizmann.ac.il Eitan Sharon

More information

The Scientific Data Mining Process

The Scientific Data Mining Process Chapter 4 The Scientific Data Mining Process When I use a word, Humpty Dumpty said, in rather a scornful tone, it means just what I choose it to mean neither more nor less. Lewis Carroll [87, p. 214] In

More information

Mean-Shift Tracking with Random Sampling

Mean-Shift Tracking with Random Sampling 1 Mean-Shift Tracking with Random Sampling Alex Po Leung, Shaogang Gong Department of Computer Science Queen Mary, University of London, London, E1 4NS Abstract In this work, boosting the efficiency of

More information

Environmental Remote Sensing GEOG 2021

Environmental Remote Sensing GEOG 2021 Environmental Remote Sensing GEOG 2021 Lecture 4 Image classification 2 Purpose categorising data data abstraction / simplification data interpretation mapping for land cover mapping use land cover class

More information

Determining optimal window size for texture feature extraction methods

Determining optimal window size for texture feature extraction methods IX Spanish Symposium on Pattern Recognition and Image Analysis, Castellon, Spain, May 2001, vol.2, 237-242, ISBN: 84-8021-351-5. Determining optimal window size for texture feature extraction methods Domènec

More information

Edge tracking for motion segmentation and depth ordering

Edge tracking for motion segmentation and depth ordering Edge tracking for motion segmentation and depth ordering P. Smith, T. Drummond and R. Cipolla Department of Engineering University of Cambridge Cambridge CB2 1PZ,UK {pas1001 twd20 cipolla}@eng.cam.ac.uk

More information

Segmentation & Clustering

Segmentation & Clustering EECS 442 Computer vision Segmentation & Clustering Segmentation in human vision K-mean clustering Mean-shift Graph-cut Reading: Chapters 14 [FP] Some slides of this lectures are courtesy of prof F. Li,

More information

Vision based Vehicle Tracking using a high angle camera

Vision based Vehicle Tracking using a high angle camera Vision based Vehicle Tracking using a high angle camera Raúl Ignacio Ramos García Dule Shu gramos@clemson.edu dshu@clemson.edu Abstract A vehicle tracking and grouping algorithm is presented in this work

More information

Speed Performance Improvement of Vehicle Blob Tracking System

Speed Performance Improvement of Vehicle Blob Tracking System Speed Performance Improvement of Vehicle Blob Tracking System Sung Chun Lee and Ram Nevatia University of Southern California, Los Angeles, CA 90089, USA sungchun@usc.edu, nevatia@usc.edu Abstract. A speed

More information

ECE 533 Project Report Ashish Dhawan Aditi R. Ganesan

ECE 533 Project Report Ashish Dhawan Aditi R. Ganesan Handwritten Signature Verification ECE 533 Project Report by Ashish Dhawan Aditi R. Ganesan Contents 1. Abstract 3. 2. Introduction 4. 3. Approach 6. 4. Pre-processing 8. 5. Feature Extraction 9. 6. Verification

More information

Tracking performance evaluation on PETS 2015 Challenge datasets

Tracking performance evaluation on PETS 2015 Challenge datasets Tracking performance evaluation on PETS 2015 Challenge datasets Tahir Nawaz, Jonathan Boyle, Longzhen Li and James Ferryman Computational Vision Group, School of Systems Engineering University of Reading,

More information

Tracking of Small Unmanned Aerial Vehicles

Tracking of Small Unmanned Aerial Vehicles Tracking of Small Unmanned Aerial Vehicles Steven Krukowski Adrien Perkins Aeronautics and Astronautics Stanford University Stanford, CA 94305 Email: spk170@stanford.edu Aeronautics and Astronautics Stanford

More information

Modelling, Extraction and Description of Intrinsic Cues of High Resolution Satellite Images: Independent Component Analysis based approaches

Modelling, Extraction and Description of Intrinsic Cues of High Resolution Satellite Images: Independent Component Analysis based approaches Modelling, Extraction and Description of Intrinsic Cues of High Resolution Satellite Images: Independent Component Analysis based approaches PhD Thesis by Payam Birjandi Director: Prof. Mihai Datcu Problematic

More information

Face Model Fitting on Low Resolution Images

Face Model Fitting on Low Resolution Images Face Model Fitting on Low Resolution Images Xiaoming Liu Peter H. Tu Frederick W. Wheeler Visualization and Computer Vision Lab General Electric Global Research Center Niskayuna, NY, 1239, USA {liux,tu,wheeler}@research.ge.com

More information

Tracking Groups of Pedestrians in Video Sequences

Tracking Groups of Pedestrians in Video Sequences Tracking Groups of Pedestrians in Video Sequences Jorge S. Marques Pedro M. Jorge Arnaldo J. Abrantes J. M. Lemos IST / ISR ISEL / IST ISEL INESC-ID / IST Lisbon, Portugal Lisbon, Portugal Lisbon, Portugal

More information

A New Robust Algorithm for Video Text Extraction

A New Robust Algorithm for Video Text Extraction A New Robust Algorithm for Video Text Extraction Pattern Recognition, vol. 36, no. 6, June 2003 Edward K. Wong and Minya Chen School of Electrical Engineering and Computer Science Kyungpook National Univ.

More information

Detection and Restoration of Vertical Non-linear Scratches in Digitized Film Sequences

Detection and Restoration of Vertical Non-linear Scratches in Digitized Film Sequences Detection and Restoration of Vertical Non-linear Scratches in Digitized Film Sequences Byoung-moon You 1, Kyung-tack Jung 2, Sang-kook Kim 2, and Doo-sung Hwang 3 1 L&Y Vision Technologies, Inc., Daejeon,

More information

A Learning Based Method for Super-Resolution of Low Resolution Images

A Learning Based Method for Super-Resolution of Low Resolution Images A Learning Based Method for Super-Resolution of Low Resolution Images Emre Ugur June 1, 2004 emre.ugur@ceng.metu.edu.tr Abstract The main objective of this project is the study of a learning based method

More information

Classification of Fingerprints. Sarat C. Dass Department of Statistics & Probability

Classification of Fingerprints. Sarat C. Dass Department of Statistics & Probability Classification of Fingerprints Sarat C. Dass Department of Statistics & Probability Fingerprint Classification Fingerprint classification is a coarse level partitioning of a fingerprint database into smaller

More information

Image Content-Based Email Spam Image Filtering

Image Content-Based Email Spam Image Filtering Image Content-Based Email Spam Image Filtering Jianyi Wang and Kazuki Katagishi Abstract With the population of Internet around the world, email has become one of the main methods of communication among

More information

PATTERN RECOGNITION AND MACHINE LEARNING CHAPTER 4: LINEAR MODELS FOR CLASSIFICATION

PATTERN RECOGNITION AND MACHINE LEARNING CHAPTER 4: LINEAR MODELS FOR CLASSIFICATION PATTERN RECOGNITION AND MACHINE LEARNING CHAPTER 4: LINEAR MODELS FOR CLASSIFICATION Introduction In the previous chapter, we explored a class of regression models having particularly simple analytical

More information

Multiscale Object-Based Classification of Satellite Images Merging Multispectral Information with Panchromatic Textural Features

Multiscale Object-Based Classification of Satellite Images Merging Multispectral Information with Panchromatic Textural Features Remote Sensing and Geoinformation Lena Halounová, Editor not only for Scientific Cooperation EARSeL, 2011 Multiscale Object-Based Classification of Satellite Images Merging Multispectral Information with

More information

Part-Based Recognition

Part-Based Recognition Part-Based Recognition Benedict Brown CS597D, Fall 2003 Princeton University CS 597D, Part-Based Recognition p. 1/32 Introduction Many objects are made up of parts It s presumably easier to identify simple

More information

Image Segmentation and Registration

Image Segmentation and Registration Image Segmentation and Registration Dr. Christine Tanner (tanner@vision.ee.ethz.ch) Computer Vision Laboratory, ETH Zürich Dr. Verena Kaynig, Machine Learning Laboratory, ETH Zürich Outline Segmentation

More information

MVA ENS Cachan. Lecture 2: Logistic regression & intro to MIL Iasonas Kokkinos Iasonas.kokkinos@ecp.fr

MVA ENS Cachan. Lecture 2: Logistic regression & intro to MIL Iasonas Kokkinos Iasonas.kokkinos@ecp.fr Machine Learning for Computer Vision 1 MVA ENS Cachan Lecture 2: Logistic regression & intro to MIL Iasonas Kokkinos Iasonas.kokkinos@ecp.fr Department of Applied Mathematics Ecole Centrale Paris Galen

More information

Circle Object Recognition Based on Monocular Vision for Home Security Robot

Circle Object Recognition Based on Monocular Vision for Home Security Robot Journal of Applied Science and Engineering, Vol. 16, No. 3, pp. 261 268 (2013) DOI: 10.6180/jase.2013.16.3.05 Circle Object Recognition Based on Monocular Vision for Home Security Robot Shih-An Li, Ching-Chang

More information

Practical Tour of Visual tracking. David Fleet and Allan Jepson January, 2006

Practical Tour of Visual tracking. David Fleet and Allan Jepson January, 2006 Practical Tour of Visual tracking David Fleet and Allan Jepson January, 2006 Designing a Visual Tracker: What is the state? pose and motion (position, velocity, acceleration, ) shape (size, deformation,

More information

Crowdclustering with Sparse Pairwise Labels: A Matrix Completion Approach

Crowdclustering with Sparse Pairwise Labels: A Matrix Completion Approach Outline Crowdclustering with Sparse Pairwise Labels: A Matrix Completion Approach Jinfeng Yi, Rong Jin, Anil K. Jain, Shaili Jain 2012 Presented By : KHALID ALKOBAYER Crowdsourcing and Crowdclustering

More information

LOCAL SURFACE PATCH BASED TIME ATTENDANCE SYSTEM USING FACE. indhubatchvsa@gmail.com

LOCAL SURFACE PATCH BASED TIME ATTENDANCE SYSTEM USING FACE. indhubatchvsa@gmail.com LOCAL SURFACE PATCH BASED TIME ATTENDANCE SYSTEM USING FACE 1 S.Manikandan, 2 S.Abirami, 2 R.Indumathi, 2 R.Nandhini, 2 T.Nanthini 1 Assistant Professor, VSA group of institution, Salem. 2 BE(ECE), VSA

More information

Low-resolution Character Recognition by Video-based Super-resolution

Low-resolution Character Recognition by Video-based Super-resolution 2009 10th International Conference on Document Analysis and Recognition Low-resolution Character Recognition by Video-based Super-resolution Ataru Ohkura 1, Daisuke Deguchi 1, Tomokazu Takahashi 2, Ichiro

More information

Segmentation of building models from dense 3D point-clouds

Segmentation of building models from dense 3D point-clouds Segmentation of building models from dense 3D point-clouds Joachim Bauer, Konrad Karner, Konrad Schindler, Andreas Klaus, Christopher Zach VRVis Research Center for Virtual Reality and Visualization, Institute

More information

DEVELOPMENT OF A SUPERVISED SOFTWARE TOOL FOR AUTOMATED DETERMINATION OF OPTIMAL SEGMENTATION PARAMETERS FOR ECOGNITION

DEVELOPMENT OF A SUPERVISED SOFTWARE TOOL FOR AUTOMATED DETERMINATION OF OPTIMAL SEGMENTATION PARAMETERS FOR ECOGNITION DEVELOPMENT OF A SUPERVISED SOFTWARE TOOL FOR AUTOMATED DETERMINATION OF OPTIMAL SEGMENTATION PARAMETERS FOR ECOGNITION Y. Zhang* a, T. Maxwell, H. Tong, V. Dey a University of New Brunswick, Geodesy &

More information

A Method of Caption Detection in News Video

A Method of Caption Detection in News Video 3rd International Conference on Multimedia Technology(ICMT 3) A Method of Caption Detection in News Video He HUANG, Ping SHI Abstract. News video is one of the most important media for people to get information.

More information

The Role of Size Normalization on the Recognition Rate of Handwritten Numerals

The Role of Size Normalization on the Recognition Rate of Handwritten Numerals The Role of Size Normalization on the Recognition Rate of Handwritten Numerals Chun Lei He, Ping Zhang, Jianxiong Dong, Ching Y. Suen, Tien D. Bui Centre for Pattern Recognition and Machine Intelligence,

More information

Vision-Based Blind Spot Detection Using Optical Flow

Vision-Based Blind Spot Detection Using Optical Flow Vision-Based Blind Spot Detection Using Optical Flow M.A. Sotelo 1, J. Barriga 1, D. Fernández 1, I. Parra 1, J.E. Naranjo 2, M. Marrón 1, S. Alvarez 1, and M. Gavilán 1 1 Department of Electronics, University

More information

FIRE AND SMOKE DETECTION WITHOUT SENSORS: IMAGE PROCESSING BASED APPROACH

FIRE AND SMOKE DETECTION WITHOUT SENSORS: IMAGE PROCESSING BASED APPROACH FIRE AND SMOKE DETECTION WITHOUT SENSORS: IMAGE PROCESSING BASED APPROACH Turgay Çelik, Hüseyin Özkaramanlı, and Hasan Demirel Electrical and Electronic Engineering, Eastern Mediterranean University Gazimağusa,

More information

Object Recognition. Selim Aksoy. Bilkent University saksoy@cs.bilkent.edu.tr

Object Recognition. Selim Aksoy. Bilkent University saksoy@cs.bilkent.edu.tr Image Classification and Object Recognition Selim Aksoy Department of Computer Engineering Bilkent University saksoy@cs.bilkent.edu.tr Image classification Image (scene) classification is a fundamental

More information

Assessment. Presenter: Yupu Zhang, Guoliang Jin, Tuo Wang Computer Vision 2008 Fall

Assessment. Presenter: Yupu Zhang, Guoliang Jin, Tuo Wang Computer Vision 2008 Fall Automatic Photo Quality Assessment Presenter: Yupu Zhang, Guoliang Jin, Tuo Wang Computer Vision 2008 Fall Estimating i the photorealism of images: Distinguishing i i paintings from photographs h Florin

More information

Class-specific Sparse Coding for Learning of Object Representations

Class-specific Sparse Coding for Learning of Object Representations Class-specific Sparse Coding for Learning of Object Representations Stephan Hasler, Heiko Wersing, and Edgar Körner Honda Research Institute Europe GmbH Carl-Legien-Str. 30, 63073 Offenbach am Main, Germany

More information

Open Access A Facial Expression Recognition Algorithm Based on Local Binary Pattern and Empirical Mode Decomposition

Open Access A Facial Expression Recognition Algorithm Based on Local Binary Pattern and Empirical Mode Decomposition Send Orders for Reprints to reprints@benthamscience.ae The Open Electrical & Electronic Engineering Journal, 2014, 8, 599-604 599 Open Access A Facial Expression Recognition Algorithm Based on Local Binary

More information

Template-based Eye and Mouth Detection for 3D Video Conferencing

Template-based Eye and Mouth Detection for 3D Video Conferencing Template-based Eye and Mouth Detection for 3D Video Conferencing Jürgen Rurainsky and Peter Eisert Fraunhofer Institute for Telecommunications - Heinrich-Hertz-Institute, Image Processing Department, Einsteinufer

More information

Feature Selection vs. Extraction

Feature Selection vs. Extraction Feature Selection In many applications, we often encounter a very large number of potential features that can be used Which subset of features should be used for the best classification? Need for a small

More information

Galaxy Morphological Classification

Galaxy Morphological Classification Galaxy Morphological Classification Jordan Duprey and James Kolano Abstract To solve the issue of galaxy morphological classification according to a classification scheme modelled off of the Hubble Sequence,

More information

Categorical Data Visualization and Clustering Using Subjective Factors

Categorical Data Visualization and Clustering Using Subjective Factors Categorical Data Visualization and Clustering Using Subjective Factors Chia-Hui Chang and Zhi-Kai Ding Department of Computer Science and Information Engineering, National Central University, Chung-Li,

More information

Automated Sudoku Solver. Marty Otzenberger EGGN 510 December 4, 2012

Automated Sudoku Solver. Marty Otzenberger EGGN 510 December 4, 2012 Automated Sudoku Solver Marty Otzenberger EGGN 510 December 4, 2012 Outline Goal Problem Elements Initial Testing Test Images Approaches Final Algorithm Results/Statistics Conclusions Problems/Limits Future

More information

Colour Image Segmentation Technique for Screen Printing

Colour Image Segmentation Technique for Screen Printing 60 R.U. Hewage and D.U.J. Sonnadara Department of Physics, University of Colombo, Sri Lanka ABSTRACT Screen-printing is an industry with a large number of applications ranging from printing mobile phone

More information

Palmprint Recognition. By Sree Rama Murthy kora Praveen Verma Yashwant Kashyap

Palmprint Recognition. By Sree Rama Murthy kora Praveen Verma Yashwant Kashyap Palmprint Recognition By Sree Rama Murthy kora Praveen Verma Yashwant Kashyap Palm print Palm Patterns are utilized in many applications: 1. To correlate palm patterns with medical disorders, e.g. genetic

More information

Ensemble Methods. Knowledge Discovery and Data Mining 2 (VU) (707.004) Roman Kern. KTI, TU Graz 2015-03-05

Ensemble Methods. Knowledge Discovery and Data Mining 2 (VU) (707.004) Roman Kern. KTI, TU Graz 2015-03-05 Ensemble Methods Knowledge Discovery and Data Mining 2 (VU) (707004) Roman Kern KTI, TU Graz 2015-03-05 Roman Kern (KTI, TU Graz) Ensemble Methods 2015-03-05 1 / 38 Outline 1 Introduction 2 Classification

More information

DYNAMIC RANGE IMPROVEMENT THROUGH MULTIPLE EXPOSURES. Mark A. Robertson, Sean Borman, and Robert L. Stevenson

DYNAMIC RANGE IMPROVEMENT THROUGH MULTIPLE EXPOSURES. Mark A. Robertson, Sean Borman, and Robert L. Stevenson c 1999 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or

More information

Bernice E. Rogowitz and Holly E. Rushmeier IBM TJ Watson Research Center, P.O. Box 704, Yorktown Heights, NY USA

Bernice E. Rogowitz and Holly E. Rushmeier IBM TJ Watson Research Center, P.O. Box 704, Yorktown Heights, NY USA Are Image Quality Metrics Adequate to Evaluate the Quality of Geometric Objects? Bernice E. Rogowitz and Holly E. Rushmeier IBM TJ Watson Research Center, P.O. Box 704, Yorktown Heights, NY USA ABSTRACT

More information

REAL TIME TRAFFIC LIGHT CONTROL USING IMAGE PROCESSING

REAL TIME TRAFFIC LIGHT CONTROL USING IMAGE PROCESSING REAL TIME TRAFFIC LIGHT CONTROL USING IMAGE PROCESSING Ms.PALLAVI CHOUDEKAR Ajay Kumar Garg Engineering College, Department of electrical and electronics Ms.SAYANTI BANERJEE Ajay Kumar Garg Engineering

More information

Data Mining: A Preprocessing Engine

Data Mining: A Preprocessing Engine Journal of Computer Science 2 (9): 735-739, 2006 ISSN 1549-3636 2005 Science Publications Data Mining: A Preprocessing Engine Luai Al Shalabi, Zyad Shaaban and Basel Kasasbeh Applied Science University,

More information

Identification of TV Programs Based on Provider s Logo Analysis

Identification of TV Programs Based on Provider s Logo Analysis AUTOMATYKA 2010 Tom 14 Zeszyt 3/1 Marta Chodyka*, W³odzimierz Mosorow** Identification of TV Programs Based on Provider s Logo Analysis 1. Introduction The problem of an easy access of underage persons

More information

Introducing diversity among the models of multi-label classification ensemble

Introducing diversity among the models of multi-label classification ensemble Introducing diversity among the models of multi-label classification ensemble Lena Chekina, Lior Rokach and Bracha Shapira Ben-Gurion University of the Negev Dept. of Information Systems Engineering and

More information

COMPARISON OF OBJECT BASED AND PIXEL BASED CLASSIFICATION OF HIGH RESOLUTION SATELLITE IMAGES USING ARTIFICIAL NEURAL NETWORKS

COMPARISON OF OBJECT BASED AND PIXEL BASED CLASSIFICATION OF HIGH RESOLUTION SATELLITE IMAGES USING ARTIFICIAL NEURAL NETWORKS COMPARISON OF OBJECT BASED AND PIXEL BASED CLASSIFICATION OF HIGH RESOLUTION SATELLITE IMAGES USING ARTIFICIAL NEURAL NETWORKS B.K. Mohan and S. N. Ladha Centre for Studies in Resources Engineering IIT

More information

86 S. Ullman, E. Sali, and M. Vidal-Naquet

86 S. Ullman, E. Sali, and M. Vidal-Naquet A Fragment-Based Approach to Object Representation and Classification Shimon Ullman, Erez Sali, and Michel Vidal-Naquet The Weizmann Institute of Science Rehvot 76100, Israel Shimon@wisdom.weizmann.ac.il

More information

Object Recognition and Template Matching

Object Recognition and Template Matching Object Recognition and Template Matching Template Matching A template is a small image (sub-image) The goal is to find occurrences of this template in a larger image That is, you want to find matches of

More information

Recognizing Cats and Dogs with Shape and Appearance based Models. Group Member: Chu Wang, Landu Jiang

Recognizing Cats and Dogs with Shape and Appearance based Models. Group Member: Chu Wang, Landu Jiang Recognizing Cats and Dogs with Shape and Appearance based Models Group Member: Chu Wang, Landu Jiang Abstract Recognizing cats and dogs from images is a challenging competition raised by Kaggle platform

More information

Binary Image Scanning Algorithm for Cane Segmentation

Binary Image Scanning Algorithm for Cane Segmentation Binary Image Scanning Algorithm for Cane Segmentation Ricardo D. C. Marin Department of Computer Science University Of Canterbury Canterbury, Christchurch ricardo.castanedamarin@pg.canterbury.ac.nz Tom

More information

3D Model based Object Class Detection in An Arbitrary View

3D Model based Object Class Detection in An Arbitrary View 3D Model based Object Class Detection in An Arbitrary View Pingkun Yan, Saad M. Khan, Mubarak Shah School of Electrical Engineering and Computer Science University of Central Florida http://www.eecs.ucf.edu/

More information

A Robust Method for Solving Transcendental Equations

A Robust Method for Solving Transcendental Equations www.ijcsi.org 413 A Robust Method for Solving Transcendental Equations Md. Golam Moazzam, Amita Chakraborty and Md. Al-Amin Bhuiyan Department of Computer Science and Engineering, Jahangirnagar University,

More information

Image Estimation Algorithm for Out of Focus and Blur Images to Retrieve the Barcode Value

Image Estimation Algorithm for Out of Focus and Blur Images to Retrieve the Barcode Value IJSTE - International Journal of Science Technology & Engineering Volume 1 Issue 10 April 2015 ISSN (online): 2349-784X Image Estimation Algorithm for Out of Focus and Blur Images to Retrieve the Barcode

More information

COLOR-BASED PRINTED CIRCUIT BOARD SOLDER SEGMENTATION

COLOR-BASED PRINTED CIRCUIT BOARD SOLDER SEGMENTATION COLOR-BASED PRINTED CIRCUIT BOARD SOLDER SEGMENTATION Tz-Sheng Peng ( 彭 志 昇 ), Chiou-Shann Fuh ( 傅 楸 善 ) Dept. of Computer Science and Information Engineering, National Taiwan University E-mail: r96922118@csie.ntu.edu.tw

More information

Using Lexical Similarity in Handwritten Word Recognition

Using Lexical Similarity in Handwritten Word Recognition Using Lexical Similarity in Handwritten Word Recognition Jaehwa Park and Venu Govindaraju Center of Excellence for Document Analysis and Recognition (CEDAR) Department of Computer Science and Engineering

More information

E27 SPRING 2013 ZUCKER PROJECT 2 PROJECT 2 AUGMENTED REALITY GAMING SYSTEM

E27 SPRING 2013 ZUCKER PROJECT 2 PROJECT 2 AUGMENTED REALITY GAMING SYSTEM PROJECT 2 AUGMENTED REALITY GAMING SYSTEM OVERVIEW For this project, you will implement the augmented reality gaming system that you began to design during Exam 1. The system consists of a computer, projector,

More information

Subspace Analysis and Optimization for AAM Based Face Alignment

Subspace Analysis and Optimization for AAM Based Face Alignment Subspace Analysis and Optimization for AAM Based Face Alignment Ming Zhao Chun Chen College of Computer Science Zhejiang University Hangzhou, 310027, P.R.China zhaoming1999@zju.edu.cn Stan Z. Li Microsoft

More information

CS 534: Computer Vision 3D Model-based recognition

CS 534: Computer Vision 3D Model-based recognition CS 534: Computer Vision 3D Model-based recognition Ahmed Elgammal Dept of Computer Science CS 534 3D Model-based Vision - 1 High Level Vision Object Recognition: What it means? Two main recognition tasks:!

More information

Data Clustering. Dec 2nd, 2013 Kyrylo Bessonov

Data Clustering. Dec 2nd, 2013 Kyrylo Bessonov Data Clustering Dec 2nd, 2013 Kyrylo Bessonov Talk outline Introduction to clustering Types of clustering Supervised Unsupervised Similarity measures Main clustering algorithms k-means Hierarchical Main

More information

Visualization of large data sets using MDS combined with LVQ.

Visualization of large data sets using MDS combined with LVQ. Visualization of large data sets using MDS combined with LVQ. Antoine Naud and Włodzisław Duch Department of Informatics, Nicholas Copernicus University, Grudziądzka 5, 87-100 Toruń, Poland. www.phys.uni.torun.pl/kmk

More information

An Energy-Based Vehicle Tracking System using Principal Component Analysis and Unsupervised ART Network

An Energy-Based Vehicle Tracking System using Principal Component Analysis and Unsupervised ART Network Proceedings of the 8th WSEAS Int. Conf. on ARTIFICIAL INTELLIGENCE, KNOWLEDGE ENGINEERING & DATA BASES (AIKED '9) ISSN: 179-519 435 ISBN: 978-96-474-51-2 An Energy-Based Vehicle Tracking System using Principal

More information

Building an Advanced Invariant Real-Time Human Tracking System

Building an Advanced Invariant Real-Time Human Tracking System UDC 004.41 Building an Advanced Invariant Real-Time Human Tracking System Fayez Idris 1, Mazen Abu_Zaher 2, Rashad J. Rasras 3, and Ibrahiem M. M. El Emary 4 1 School of Informatics and Computing, German-Jordanian

More information

FACE RECOGNITION BASED ATTENDANCE MARKING SYSTEM

FACE RECOGNITION BASED ATTENDANCE MARKING SYSTEM Available Online at www.ijcsmc.com International Journal of Computer Science and Mobile Computing A Monthly Journal of Computer Science and Information Technology IJCSMC, Vol. 3, Issue. 2, February 2014,

More information

Proc. First Users Conference of the National Library of Medicine's Visible Human Project, active contours free forms Triangle meshes may be imp

Proc. First Users Conference of the National Library of Medicine's Visible Human Project, active contours free forms Triangle meshes may be imp Proc. First Users Conference of the National Library of Medicine's Visible Human Project, 1996 1 Creation of Finite Element Models of Human Body based upon Tissue Classied Voxel Representations M. Muller,

More information

Make and Model Recognition of Cars

Make and Model Recognition of Cars Make and Model Recognition of Cars Sparta Cheung Department of Electrical and Computer Engineering University of California, San Diego 9500 Gilman Dr., La Jolla, CA 92093 sparta@ucsd.edu Alice Chu Department

More information

HANDS-FREE PC CONTROL CONTROLLING OF MOUSE CURSOR USING EYE MOVEMENT

HANDS-FREE PC CONTROL CONTROLLING OF MOUSE CURSOR USING EYE MOVEMENT International Journal of Scientific and Research Publications, Volume 2, Issue 4, April 2012 1 HANDS-FREE PC CONTROL CONTROLLING OF MOUSE CURSOR USING EYE MOVEMENT Akhil Gupta, Akash Rathi, Dr. Y. Radhika

More information

Extraction of Satellite Image using Particle Swarm Optimization

Extraction of Satellite Image using Particle Swarm Optimization Extraction of Satellite Image using Particle Swarm Optimization Er.Harish Kundra Assistant Professor & Head Rayat Institute of Engineering & IT, Railmajra, Punjab,India. Dr. V.K.Panchal Director, DTRL,DRDO,

More information

MANAGING QUEUE STABILITY USING ART2 IN ACTIVE QUEUE MANAGEMENT FOR CONGESTION CONTROL

MANAGING QUEUE STABILITY USING ART2 IN ACTIVE QUEUE MANAGEMENT FOR CONGESTION CONTROL MANAGING QUEUE STABILITY USING ART2 IN ACTIVE QUEUE MANAGEMENT FOR CONGESTION CONTROL G. Maria Priscilla 1 and C. P. Sumathi 2 1 S.N.R. Sons College (Autonomous), Coimbatore, India 2 SDNB Vaishnav College

More information

An Experimental Study of the Performance of Histogram Equalization for Image Enhancement

An Experimental Study of the Performance of Histogram Equalization for Image Enhancement International Journal of Computer Sciences and Engineering Open Access Research Paper Volume-4, Special Issue-2, April 216 E-ISSN: 2347-2693 An Experimental Study of the Performance of Histogram Equalization

More information

Potential of face area data for predicting sharpness of natural images

Potential of face area data for predicting sharpness of natural images Potential of face area data for predicting sharpness of natural images Mikko Nuutinen a, Olli Orenius b, Timo Säämänen b, Pirkko Oittinen a a Dept. of Media Technology, Aalto University School of Science

More information

Determining the Resolution of Scanned Document Images

Determining the Resolution of Scanned Document Images Presented at IS&T/SPIE EI 99, Conference 3651, Document Recognition and Retrieval VI Jan 26-28, 1999, San Jose, CA. Determining the Resolution of Scanned Document Images Dan S. Bloomberg Xerox Palo Alto

More information

DATA MINING CLUSTER ANALYSIS: BASIC CONCEPTS

DATA MINING CLUSTER ANALYSIS: BASIC CONCEPTS DATA MINING CLUSTER ANALYSIS: BASIC CONCEPTS 1 AND ALGORITHMS Chiara Renso KDD-LAB ISTI- CNR, Pisa, Italy WHAT IS CLUSTER ANALYSIS? Finding groups of objects such that the objects in a group will be similar

More information

Augmented Reality Tic-Tac-Toe

Augmented Reality Tic-Tac-Toe Augmented Reality Tic-Tac-Toe Joe Maguire, David Saltzman Department of Electrical Engineering jmaguire@stanford.edu, dsaltz@stanford.edu Abstract: This project implements an augmented reality version

More information

Local features and matching. Image classification & object localization

Local features and matching. Image classification & object localization Overview Instance level search Local features and matching Efficient visual recognition Image classification & object localization Category recognition Image classification: assigning a class label to

More information

More Local Structure Information for Make-Model Recognition

More Local Structure Information for Make-Model Recognition More Local Structure Information for Make-Model Recognition David Anthony Torres Dept. of Computer Science The University of California at San Diego La Jolla, CA 9093 Abstract An object classification

More information

MINITAB ASSISTANT WHITE PAPER

MINITAB ASSISTANT WHITE PAPER MINITAB ASSISTANT WHITE PAPER This paper explains the research conducted by Minitab statisticians to develop the methods and data checks used in the Assistant in Minitab 17 Statistical Software. One-Way

More information

CS231M Project Report - Automated Real-Time Face Tracking and Blending

CS231M Project Report - Automated Real-Time Face Tracking and Blending CS231M Project Report - Automated Real-Time Face Tracking and Blending Steven Lee, slee2010@stanford.edu June 6, 2015 1 Introduction Summary statement: The goal of this project is to create an Android

More information

A Reliability Point and Kalman Filter-based Vehicle Tracking Technique

A Reliability Point and Kalman Filter-based Vehicle Tracking Technique A Reliability Point and Kalman Filter-based Vehicle Tracing Technique Soo Siang Teoh and Thomas Bräunl Abstract This paper introduces a technique for tracing the movement of vehicles in consecutive video

More information

Handwritten Character Recognition from Bank Cheque

Handwritten Character Recognition from Bank Cheque International Journal of Computer Sciences and Engineering Open Access Research Paper Volume-4, Special Issue-1 E-ISSN: 2347-2693 Handwritten Character Recognition from Bank Cheque Siddhartha Banerjee*

More information

A Study on SURF Algorithm and Real-Time Tracking Objects Using Optical Flow

A Study on SURF Algorithm and Real-Time Tracking Objects Using Optical Flow , pp.233-237 http://dx.doi.org/10.14257/astl.2014.51.53 A Study on SURF Algorithm and Real-Time Tracking Objects Using Optical Flow Giwoo Kim 1, Hye-Youn Lim 1 and Dae-Seong Kang 1, 1 Department of electronices

More information

An Optical Sudoku Solver

An Optical Sudoku Solver An Optical Sudoku Solver Martin Byröd February 12, 07 Abstract In this report, a vision-based sudoku solver is described. The solver is capable of solving a sudoku directly from a photograph taken with

More information

Data Preprocessing. Week 2

Data Preprocessing. Week 2 Data Preprocessing Week 2 Topics Data Types Data Repositories Data Preprocessing Present homework assignment #1 Team Homework Assignment #2 Read pp. 227 240, pp. 250 250, and pp. 259 263 the text book.

More information

suggestive contours and abstracted shading Daniel Arias

suggestive contours and abstracted shading Daniel Arias suggestive contours and abstracted shading Daniel Arias LINES AND SHADES Shape, volume, shades and texture in drawing lines as nonphotorealistic rendering technique Luis Caballero Suggestive contours Many

More information

Feature Tracking and Optical Flow

Feature Tracking and Optical Flow 02/09/12 Feature Tracking and Optical Flow Computer Vision CS 543 / ECE 549 University of Illinois Derek Hoiem Many slides adapted from Lana Lazebnik, Silvio Saverse, who in turn adapted slides from Steve

More information

Social Media Mining. Data Mining Essentials

Social Media Mining. Data Mining Essentials Introduction Data production rate has been increased dramatically (Big Data) and we are able store much more data than before E.g., purchase data, social media data, mobile phone data Businesses and customers

More information

Such As Statements, Kindergarten Grade 8

Such As Statements, Kindergarten Grade 8 Such As Statements, Kindergarten Grade 8 This document contains the such as statements that were included in the review committees final recommendations for revisions to the mathematics Texas Essential

More information

Predicting the Stock Market with News Articles

Predicting the Stock Market with News Articles Predicting the Stock Market with News Articles Kari Lee and Ryan Timmons CS224N Final Project Introduction Stock market prediction is an area of extreme importance to an entire industry. Stock price is

More information

CHOOSING A COLLEGE. Teacher s Guide Getting Started. Nathan N. Alexander Charlotte, NC

CHOOSING A COLLEGE. Teacher s Guide Getting Started. Nathan N. Alexander Charlotte, NC Teacher s Guide Getting Started Nathan N. Alexander Charlotte, NC Purpose In this two-day lesson, students determine their best-matched college. They use decision-making strategies based on their preferences

More information

GameTime: A Toolkit for Timing Analysis of Software

GameTime: A Toolkit for Timing Analysis of Software GameTime: A Toolkit for Timing Analysis of Software Sanjit A. Seshia and Jonathan Kotker EECS Department, UC Berkeley {sseshia,jamhoot}@eecs.berkeley.edu Abstract. Timing analysis is a key step in the

More information

Support Vector Machines with Clustering for Training with Very Large Datasets

Support Vector Machines with Clustering for Training with Very Large Datasets Support Vector Machines with Clustering for Training with Very Large Datasets Theodoros Evgeniou Technology Management INSEAD Bd de Constance, Fontainebleau 77300, France theodoros.evgeniou@insead.fr Massimiliano

More information